How does the amount of nm (32/45) matter?

[ricno]

I have a question regarding CPUs. Often when looking at a certain processor one of the main things noted from Intel is the number of nm. (First question is really if that is nano meter?) That could be for example 32 nm, 45 or 65 nm.)

And the second question is, why does this matter? Or how does it relate to other things?

Is it better for the CPU to have a lower or higher nm? Is it connected to support on the motherboard socket, or heat produced or more advanced features?

 

[buzznut]

nm is a nanometer, it is referencing the space between the semiconductor layers without getting too techinical. This has been getting smaller for years and they keep finding ways to decreases the size, but with current materials there may be limits to how small this space can be.

So it basically means that manufacturers are able to squeeze more trnsistor circuits ionto thew same space. This is particularly important to video cards, which now have as many as 3 billions transistors on one chip.

What does it mean to you? Well, the smaller size can be an indication of lower power usage and more efficient design. It may be able to provide a better overclock if it runs cooler than previous designs.

The 32nm Intel CPU's are the best design available today for the home PC. AMD is still using a 45nm design and may not reach the same efficiency as Intel until they can move to 32nm. Meanwhile Intel will be moving to an even smaller design before long.

Ultimately it isn't more important than number of cores, clock speed, or amount of cache available to the processor.

 

[ricno]

Thank you for a very good reply.

What does it mean to you? Well, the smaller size can be an indication of lower power usage and more efficient design. It may be able to provide a better overclock if it runs cooler than previous designs.

Do you know what the benefits would be for server CPUs? (Xeon/Opteron) Since no overclocking is normaly done, would it be the lower power usage that should be interesting?

AMD is still using a 45nm design and may not reach the same efficiency as Intel until they can move to 32nm.

Is there anything fixed with these specific values? As in, does AMD have to go to 32 and not e.g. 29?

 

[unknown_13]

Thank you for a very good reply.



Do you know what the benefits would be for server CPUs? (Xeon/Opteron) Since no overclocking is normaly done, would it be the lower power usage that should be interesting?



Is there anything fixed with these specific values? As in, does AMD have to go to 32 and not e.g. 29?

AMD will move to 32nm in 2011. The next step will be 22nm, sometime in late 2011/2012.

 

[ricno]

The next step will be 22nm, sometime in late 2011/2012.

That 22nm, will that be for Intel or for AMD too?

 

[unknown_13]

That 22nm, will that be for Intel or for AMD too?

Probably Intel is going to be the first of the two (Intel and AMD) to use 22nm tech.

 

[hunter315]

Is there anything fixed with these specific values? As in, does AMD have to go to 32 and not e.g. 29?

They usually decrease to 0.7 times the previous size, old P4s were 130nm, then we went to 90nm, then 65, 45, 32, 22 etc. This refers to the length of the channel in the transistor and by reducing it by 0.7, since you have them arranged in a 2 dimensional grid you can fit twice as many in the same space by reducing their size, you also get to drop their voltages and reduce their power consumption, they usually stick to 0.7 since those let them double their amount on a chip, theoretically they could drop by 0.6 but they rely on the fab sites also advancing their production equipment to be able to handle that size as well so they usually stick to standard steps since neither Intel, AMD/ATI, or nVidia own their own fabs so they are relying on shared fabs and it helps out everyone if the fab just gets to work on going to 32nm or 22nm rather than 32nm for intel, and 29nm for AMD.

Graphics cards are made in a different way so they arent running on the same sizes as the CPU fabs are, right now they are all at 40nm, and NI line up from ATI is due to be 28nm.

 

[BadTrip]

They usually decrease to 0.7 times the previous size, old P4s were 130nm, then we went to 90nm, then 65, 45, 32, 22 etc. This refers to the length of the channel in the transistor and by reducing it by 0.7, since you have them arranged in a 2 dimensional grid you can fit twice as many in the same space by reducing their size, you also get to drop their voltages and reduce their power consumption, they usually stick to 0.7 since those let them double their amount on a chip, theoretically they could drop by 0.6 but they rely on the fab sites also advancing their production equipment to be able to handle that size as well so they usually stick to standard steps since neither Intel, AMD/ATI, or nVidia own their own fabs so they are relying on shared fabs and it helps out everyone if the fab just gets to work on going to 32nm or 22nm rather than 32nm for intel, and 29nm for AMD.

Graphics cards are made in a different way so they arent running on the same sizes as the CPU fabs are, right now they are all at 40nm, and NI line up from ATI is due to be 28nm.

Intel most certianly does own it's fabs.

 

[Timop]

Intel most certianly does own it's fabs.

Yup.
Intel owns its own fabs; AMD used to, but spun it off to form GloFo; Nvidia uses TSMC; and ATI is considering switching from TSMC to GloFo for 28nm.

 

[BadTrip]

Intel was planning on using TSMC for some SOC atoms, but that agreement is on hold.

 

[rohn_avni]

why u guys always go offtopic???

 

[rohn_avni]

in simple word's lower the fabrication process lower the temps...

 

[ricno]

Best answer selected by ricno.